Pricing research with a research agent is an AI-led approach in which an agent turns a pricing question into a study design, selects the method, builds the survey, fields it, and models the results. Agents typically choose Van Westendorp for an acceptable price range, Gabor-Granger for demand at set price points, and conjoint for price trade-offs against features or competitors
Summary:
|
Price is the fastest lever most brands have. McKinsey's classic analysis of pricing found that, with volumes stable, a 1% price increase lifts operating profit by about 8% for the average S&P 1500 company. The same logic cuts the other way: a small pricing mistake costs more than a small cost or volume miss.
That is why the choice of pricing method matters. This guide shows how a research agent chooses between Van Westendorp, Gabor-Granger, and conjoint, then designs and fields the study. If your team works in consumer insights, it also shows where researchers stay in control.
What Pricing Research With a Research Agent Involves
Pricing research measures how buyers respond to price so a team can choose one with evidence. With a research agent, the agent handles the workflow end to end:
Take the pricing question and form a hypothesis.
Choose the method.
Build the survey and stimulus.
Field the study.
Model the results.
Draft a price recommendation.
An AI research agent makes these decisions in context, with a researcher approving the key ones. That separates it from automated pricing tools, which run one fixed method whatever the question. A fixed tool can only answer the question it was built for.
Adoption of AI is already broad. A McKinsey survey of 1,491 participants in July 2024 found 71% of organizations using gen AI in at least one business function.
For the quantitative end of the job, see how AI agents for MaxDiff and conjoint studies design and model trade-off tasks.
Step 1: Turning the Business Question Into a Pricing Hypothesis
The agent first classifies the decision. Is this a launch price, a price increase, a tier or pack-price architecture, a competitive response, or a promotion? Each one points to a different method.
Next it defines the output the decision needs: a price range, a single price point, a demand curve, or expected market share. A team deciding between three price points needs something different from a team that has no idea where to start.
Finally, it sets constraints: budget, timeline, sample, and whether the product is a concept or already in market. A concept has no reference price, while an in-market product does.
Bain's January 2025 Commercial Excellence Survey shows why evidence matters here. Among 1,263 respondents, 67% named competitive pressure and customer resistance as the biggest barrier to price increases, and 39% cited insufficient data or analytics. A well-designed study supplies the customer evidence that pricing teams say they lack.
Step 2: How the Agent Chooses the Pricing Method
Method choice rests on five criteria:
Output needed: a range, a demand curve, or share and trade-offs.
Competitive context: whether buyers compare alternatives when they decide.
Feature trade-offs: whether price interacts with features or tiers.
Sample and time: what the budget and schedule allow.
Buyer familiarity: whether buyers know the product category and typical prices.
Here is the decision as a simple tree:
No reference price, early concept? Run Van Westendorp for an acceptable range.
Known candidate price points, or testing a price increase? Run Gabor-Granger for demand and revenue curves.
Price depends on competitors, features, or tiers? Run conjoint with a market simulator.
One realistic reaction to a single price, or the premium over a rival? Run monadic price testing or brand-price trade-off.
Pricing decision | Output needed | Typical method |
Early concept price discovery | Acceptable price range | Van Westendorp |
Launch price among known options | Demand and revenue curves | Gabor-Granger |
Price increase | Demand at higher price points | Gabor-Granger or monadic |
Tier or feature-based pricing | Utilities and share | Conjoint |
Competitive response | Switching and premium | Brand-price trade-off or conjoint |
Teams that want to understand how buyers respond to price changes can start with the guide to price sensitivity.
When the Agent Selects Van Westendorp
Van Westendorp suits early concepts with no reference price, where the goal is an acceptable price range. It asks respondents where a product becomes too cheap or too expensive, and the answers define a range for early-stage price discovery. The existing guide to Van Westendorp pricing covers the mechanics.
The method has limits. It shows where buyers draw lines, not how many would buy at a given price. When that gap matters, the agent adds a purchase intent calibration question, so the range can be tied to likely demand.
When the Agent Selects Gabor-Granger
Gabor-Granger suits cases with known candidate price points, including testing a price increase. Respondents answer whether they would buy at a series of prices, and the answers build a demand curve and a revenue-maximizing price. Our guide to the Gabor-Granger pricing method explains the calculation.
The main risk is anchoring: the first price a respondent sees colors every answer after it. The agent randomizes the starting price across respondents to reduce that effect. Anchoring is one of several cognitive biases that can distort research results.
When the Agent Selects Conjoint
Conjoint suits pricing that depends on competitors, features, or tier design. Respondents choose between full product profiles, so the model estimates part-worth utilities for each feature and price level, then a market simulator predicts share under different scenarios. The conjoint analysis guide covers the foundations.
Conjoint needs more sample, more time, and more design effort. It is worth it when price-feature trade-offs drive the decision, such as building a good-better-best tier architecture. It is not worth it when a team only needs a ballpark range.
Teams that want an on-platform route can explore Decode's conjoint testing.
When the Agent Selects Monadic Price Testing or Brand-Price Trade-Off
In monadic cells, each respondent sees one price and reacts to it without comparison. That gives a realistic reaction to a single price, much like a shopper meeting it on shelf. The trade-off is sample: each person answers only one price, so monadic designs need more respondents for a clean read. Our guide to price testing for brands covers the approach.
Brand-price trade-off measures the premium a brand can command over competitors. The market context is clear in NIQ's 2025 global report, based on over 17,000 consumers in 25 countries, which found that branded products sell at a 26% premium over private label across CPG categories globally. The same report found 69% of respondents see private labels as good value, so the premium has to be earned.
Step 3: Combining Methods in One Pricing Study
Many questions need more than one method. Common pairings include:
A Van Westendorp range followed by Gabor-Granger points within it. The first narrows the field and the second picks the price.
Conjoint plus a range check for a new category. Conjoint models trade-offs, while the range check confirms the prices tested are plausible.
Module order matters. A price shown in an early module can anchor answers in the next, so the agent places range questions before specific price points, or splits respondents across modules. It also keeps competitor prices out of earlier modules unless the design calls for them, and it records why it chose each pairing so a researcher can see the logic behind the design. Decode supports this kind of design in its price testing workflows.
Step 4: Building the Pricing Survey and Stimulus
A good pricing survey starts with the concept description and value framing, shown before any price. Buyers cannot judge a price without knowing what they get, and sloppy framing moves answers more than the price itself.
The agent then sets the price ladder: the price points, their spacing, and the order controls. A ladder that is too narrow shows nothing, while one that is too wide includes prices no buyer would see. Order bias is controlled by randomizing task sequence and rotating price levels across respondents. It adds attention checks to remove inattentive respondents.
Sample comes last. The agent sets the sample size per cell and quotas, such as category buyers, current customers, and prospects. Cells that are too small cannot be read, a risk covered in types of sampling biases.
Adding Adaptive Follow-Up Questions on Price
A scale tells you a price is too high. It does not tell you why. The agent probes the reasons behind answers like "too expensive" or "would not buy", asking what the buyer expected for the price and what they would compare it to.
Open-ended reasons are then coded into value drivers by segment, such as quality, convenience, or trust. Keeping the verbatims, curves, and segments together in one consumer insights platform makes it easier to explain why a price works for one group and not another.
Step 5: Modeling Results and Making a Price Recommendation
Each method produces a different output: a range from Van Westendorp, demand and revenue curves from Gabor-Granger, and utilities and share from conjoint. The agent models them and checks how price response differs by segment, since one price rarely fits everyone.
Segment-level pricing deserves its own read. Current customers, prospects, and category buyers often respond differently, and a price that wins overall can lose a key segment. The agent reports price response by segment so the team can decide whether one price or several is right.
The recommendation comes with scenarios, margin impact, and a confidence level. A researcher should see, for example, the expected volume at each price, what it does to margin, and how sure the model is. Teams evaluating consumer research platforms should check that the platform shows this uncertainty, not just a single number.
Checking Results Against Real-World Signals
Stated answers are not purchases. The agent compares the results against sales history, competitor shelf prices, and in-market tests, then flags gaps between stated and actual buying. This is the stated versus revealed preference problem, explored in the say-do gap in CPG research.
Calibration closes the gap. If the study overstates demand for a product already in market, the agent can scale the stated results to what sales data shows, then apply the same adjustment to the new price.
Pricing Study Design Mistakes the Agent Should Catch
Common errors include:
A method that does not fit the decision: using a range method for a competitive switching question.
Unrealistic price ladders: points no buyer would see, or spacing too wide to show a pattern.
Missing competitive context: testing a price without the alternatives a shopper would compare it to.
Cells too small to read: results too noisy to act on.
Relying on stated purchase intent alone: a number that overstates demand.
Automation raises the stakes for catching these early. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value, or weak risk controls. A pricing agent needs checks built in.
Where Researchers Stay in Control
Researchers approve the method choice, the price points, and the margin floor. They interpret segment differences and own the final pricing decision. The agent proposes, and the researcher decides.
That division is covered in human-in-the-loop AI-moderated research.
Researchers should also check the agent's outputs for errors, such as a curve that does not match the raw data or a segment claim without support. Our guide on how to detect hallucinations in AI research lists the checks.
Using Price-Reveal Reactions to Break Ties Between Methods
Sometimes two methods point to slightly different prices. Reactions at the moment of price reveal can break the tie. Facial coding at the instant a price appears shows acceptance or resistance before the shopper answers, so a price that looks acceptable on a scale but triggers a negative reaction deserves a second look.
Eye tracking on shelf price tags and pricing pages shows whether price or value cues get noticed first. If shoppers see the price before the benefits, framing may need work. Treat these signals as tie-breakers, not replacements: they add context to the method result without overriding it.
Decode by Entropik supports agent-led pricing studies with this measurement. It is trusted by 150+ global brands and backed by 17 patents.
Facial coding reaches 90%+ accuracy across 62 facial expressions.
Eye tracking reaches 96% accuracy, and 70+ languages are supported for multi-market pricing.
Frequently Asked Questions
1. How does a research agent choose a pricing research method?
It classifies the decision, defines the output needed, and checks constraints. It then maps the decision to a method: Van Westendorp for a range, Gabor-Granger for demand at set prices, and conjoint for trade-offs. A researcher approves the choice.
2. What is the best pricing research method for a new product?
For an early concept with no reference price, Van Westendorp is a common starting point because it finds an acceptable range. Many teams then run Gabor-Granger within that range to choose the price.
3. Can you combine Van Westendorp and Gabor-Granger in one pricing study?
Yes. Run Van Westendorp first to set the range, then use Gabor-Granger to test points inside it. Control module order so the first does not anchor the second.
4. When is conjoint analysis worth the extra cost in pricing research?
When price depends on features, competitors, or tier design, and the decision justifies more sample, time, and design effort. For a simple range check, it is more than you need.
4. How many respondents do you need for a pricing study?
It depends on the method, the number of cells, and how precise the answer must be. Set the sample size per cell by the decision at stake and confirm it with your statistician.
5. How long does a pricing study take with a research agent?
Agents shorten design, fielding setup, and modeling. Total time still depends on recruitment and the method, and conjoint takes longer than a range or demand study.
6. How do you test a price increase without anchoring respondents?
Randomize the starting price, avoid showing the current price first unless the design requires it, and consider monadic cells where each respondent sees one price.
7. Can a research agent recommend a final price?
It can recommend a price with scenarios, margin impact, and a confidence level. The final decision belongs to the researcher and the business, who weigh strategy and risk.
Match the Pricing Method to the Decision
A pricing study should fit the decision, from method choice to a price recommendation, with researchers approving each key step.
Request a Demo to see Decode run agent-led pricing studies with facial coding at 90%+ accuracy, eye tracking at 96% accuracy, and 70+ languages.
